The service is provided by the Microsoft Azure Speech STT cloud using the HTTP protocol. \ Has no technical performance limitations.

Speech Recognition Component (Azure Cloud Speech)

:icons:font :hardbreaks: :imagesdir: ../../../assets/images :toc-title: Contents :toc:

Description

Recognizes the subscriber’s speech into text form. +\

For the procedure for registering an MS Azure account and the cost of the service, see https://learn.microsoft.com/en-us/azure/ai-services/speech-service/.

Tokens are configured in the domain settings section, field 'azure_cloud'..

Allows you to interrupt in silence after a spoken phrase.

Table 1. System characteristics

Index

231

Short name

asr_azure

Script types

IVR

Start module

r_sip_ivr_script_component_asr_azure

Mode

Asynchronous

Icon

231

Branching type

Branching, interrupting

Properties

Table 2. Properties
Specification Description

Name: Azure Account

Code: accountKey

Visibility condition: no

Default: default

Indication of the Azure account whose options are taken to connect to the cloud.

The list contains the value 'default', which sets the root field 'speech' to be used in the 'settings.azure_cloud' object.

Additionally, the list includes the keys of the 'settings.azure_cloud.accounts' object, behind each of which there is also an object with separately configured cloud access parameters.

Name: Language

Code: lang

Visibility condition: no

Default: English

Azure Speech recognition service parameter: recognition language.

Possible options:

  • Auto (100) – Automatic language detection

  • de-DE (1) – German

  • en-US (2) – English

  • es-ES (3) – Spanish

  • fi-FI (4) – Finnish

  • fr-FR (5) – French

  • he-HE (6) – Hebrew

  • it-IT (7) – Italian

  • kk-KZ (8) – Kazakh

  • nl-NL (9) – Dutch

  • pl-PL (10) – Polish

  • pt-PT (11) – Portuguese

  • pt-BR (12) – Brazilian Portuguese

  • ru-RU (13) – Russian

  • sv-SE (14) – Swedish

  • tr-TR (15) – Turkish

  • other (100) – Other

Name: Other language

Code: langCustom

Visibility condition: yes

Default: en-US

Language code in format ISO 639-1 from the list supported.

Name: Profanity filter Code: profanityFilter Visibility condition: no

Default: Disable

Azure Speech ASR recognition service parameter: profanity filter operating mode.

Possible options:

  • raw (0) - Leave profanity

  • masked (1) - Mask profanity

  • removed (2) - Remove profanity

Name: CID (custom model enpoint id) Code: cid

Visibility condition: no

Default: empty

When you’re using the Speech Studio to create custom models, you can take advantage of the Endpoint ID value from the Deployment page. Use the Endpoint ID value as the argument to the cid query string parameter.

Name: Pronunciation scores Code: pronunciation

Visibility condition: no

Default: false

Specifies the parameters for showing pronunciation scores in recognition results. These scores assess the pronunciation quality of speech input, with indicators like accuracy, fluency, and completeness.

Possible options:

  • false (0) - No. The answer does not contain pronunciation scores.

  • true (1) - Yes. In response, pronunciation scores are included, and options for customization open below.

Name: ReferenceText Code: referenceText

Visibility condition: yes

Default: empty

The text that the pronunciation is evaluated against.

Applies only if the mode for displaying information with pronunciation scores in the response is enabled.

Name: GradingSystem Code: gradingSystem

Visibility condition: yes

Default: fivepoint

The point system for score calibration.

Applies only if the mode for displaying information with pronunciation scores in the response is enabled.

Possible options:

  • fivepoint (0) - FivePoint. Gives 0-5 floating point score

  • hundreedmark (1) - HundreedMark. Gives 0-100 floating point score

Name: Granularity Code: granularity

Visibility condition: yes

Default: phoneme

The evaluation granularity.

Applies only if the mode for displaying information with pronunciation scores in the response is enabled.

Possible options:

  • phoneme (0) - Phoneme. Shows the score on the full-text, word, and phoneme levels.

  • word (1) - Word. Shows the score on the full-text and word levels.

  • fulltext (2) - FullText. Shows the score on the full-text level only.

Name: Dimension Code: dimension

Visibility condition: yes Default: basic

Defines the output criteria. Applies only if the mode for displaying information with pronunciation scores in the response is enabled.

Possible options:

  • basic (0) - Basic. Shows the accuracy score only.

  • comprehensive (1) - Comprehensive. Shows scores on more dimensions (for example, fluency score and completeness score on the full-text level, and error type on the word level).

Name: EnableMiscue Code: enableMiscue

Visibility condition: yes

Default: false

Enables miscue calculation. With this parameter enabled.

Applies only if the mode for displaying information with pronunciation scores in the response is enabled.

Possible options:

  • false (0) - False. Matching to reference text is not used.

  • true (1) - True. The pronounced words are compared to the reference text. They are marked with omission or insertion based on the comparison.

Name: ScenarioId Code: scenarioId

Visibility condition: yes

Default: __empty

A GUID that indicates a customized point system.

Applies only if the mode for displaying information with pronunciation scores in the response is enabled.

Name: Recording timeout, s

Code: recordTimeoutSec

Visibility condition: no Default: 30

Maximum allowed recording time from the end of preview playback, in seconds.

Name: Abort on DTMF

Code: checkDTMF

Visibility condition: no

Default: no

DTMF detector switch. Opens the settings for the modes of saving characters and interrupting the operation.

Name: Buffer for DTMF Code: dtmfBuffer Visibility condition: yes

Default: — 

Variable for storing received DTMF characters.

Name: Clear DTMF buffer

Code: clearDtmfBuffer

Visibility condition: yes

Default: Yes

DTMF buffer pre-clear switch.

Name: Number of characters

Code: maxSymbolCount

Visibility condition: yes

Default: — 

An argument containing a limit on the number of characters entered.

After receiving the specified number of DTMF characters during the execution of a component, the recording is automatically completed and the last portion of voice data is sent to the recognition service.

Name: Interrupt characters Code: interruptSymbols Visibility condition: yes

Default: — 

A string containing sequences of break characters separated by commas.

After detecting a character sequence at the end of the DTMF buffer that matches one of the specified interrupt sequences, the recording is automatically completed and the last portion of data is sent to the recognition service.

For example, *, 7, 123, 9395.

Name: Abort when silence is detected Code: abortOnSilence Visibility condition: no

Default: Yes

Voice detector (VAD) switch to automatically complete recording and send the last portion of voice data to the recognition service.

The criterion for stopping is the presence of voice for at least 300 ms and its subsequent absence during the specified interval.

Name: Silence interval, s

Code: silenceTimeoutSec

Visibility condition: yes Default: 2

Argument containing the timeout time when silence is detected in seconds. If a voice is heard above the sensitivity threshold within the specified timeout, the timer is reset until silence is detected again.

In most cases, 2 seconds is a comfortable interval.

Can be specified in decimal format, for example '1.6'.

Name: VAD Threshold, -dB

Code: vadThreshold

Visibility condition: yes Default: 30

Argument setting the VAD threshold. VAD detector parameter.

The specified value is used with a negative sign (in decibels) to determine the silence cutoff threshold relative to the possible maximum.

The stream is analyzed in 20 ms frames. Each audio frame is considered either voice or silence (the average loudness level of the frame) based on a cutoff threshold.

Thus, the detector has two states: voice and silence. When transitioning between states, an event is generated about the transition to another state. Initially, the detector is in a silent state.

The generation of a voice detection signal occurs in a state of silence when several frames in a row are voice (parameter 'Voice threshold, ms'). The generation of a silence detection signal occurs in the voice state after several frames in a row are recognized as silence (the 'Silence threshold, ms' parameter).

With a value of 30, the detector better detects silence; with a value of 40, the detector better detects the appearance of a voice.

The component exits after the timeout specified in the Silence interval, s field has expired, counting from the beginning of the silence segment detected by the detector from the voice state.

Name: Voice, min. duration, ms

Code: vadVoiceMinMs

Visibility condition: yes Default: 120

Argument that sets the minimum threshold VAD duration for generating the VAD UP signal.

VAD detector parameter.

To generate a signal, several 20 ms frames in a row are required with an average level above the set VAD threshold.

The specified value determines the number of frames.

Name: Voice, tolerance, % Code: vadVoiceTol Visibility condition: yes

Default: 20

An argument that sets the permissible deviation in the sequence of voice frames.

VAD detector parameter.

For example, the minimum duration is set to 210 (ms), deviation 20 (%). Thus, signaling the presence of voice would require a sequence of 11 frames, 9 of which would be voice frames.

A value of 0 means that all 11 frames in a row must be voice (have an average level above the VAD threshold).

Name: Silence, min. duration, ms

Code: vadSilenceMinMs

Visibility condition: yes

Default: 500

An argument that sets the minimum threshold duration of frames with silence for generating the VAD DOWN signal.

VAD detector parameter.

To generate a signal, several 20 ms frames in a row are required with an average level below the set VAD threshold. The specified value determines the number of frames.

Name: Silence, tolerance, %

Code: vadSilenceTol

Visibility condition: yes

Default: 10

An argument that sets the permissible deviation in the sequence of voice frames.

VAD detector parameter. For example, the minimum duration is set to 210 ms, deviation 20%. Thus, signaling the presence of voice would require a sequence of 11 frames, 9 of which would be voice frames.

A value of 0 means that all 11 frames in a row must be silent (have an average level below the VAD threshold).

Name: Response waiting timeout, s

Code: responseTimeoutSec

Visibility condition: no

Default: 5

Timeout waiting for a response from the Azure Speech recognition service after sending it the last portion of voice data. After the timeout expires, control is transferred to the next component along the Time branch.

Name: Result into variable

Code: varText

Visibility condition: no

Default: — 

Variable for saving the text recognition result.

Name: Response code to variable Code: varHttpCode Visibility condition: no

Default: — 

Variable for storing the HTTP response code of the recognition service.

Name: Response body to variable

Code: varHttpBody

Visibility condition: no

Default: — 

Variable for storing the full contents of the HTTP response of the recognition service.

Name: Save recording file

Code: saveRec

Visibility condition: no

Default: No

Switch to save the recording file sent to the recognition service.

Name: Path to file in variable

Code: varRecordPath

Visibility condition: yes Default: — 

Variable for saving the path to the recording file.

The file is located in the script’s temporary directory and will be deleted when the script completes.

To save a file for a long time, the script requires further movement of the file to a permanent storage location.

The recording is carried out on a server with the mg role, serving the current call, and then transferred to a server with the ivr role, serving the current scenario. The transfer always takes place within the site.

Name: Preview Code: prePlayFile Visibility condition: no

Default: — 

Pre-play audio file for the subscriber, during which the voice detector is also activated.

If there is no voice from the subscriber (taking into account the noise threshold of the VAD detector), data is not sent to the recognition service.

Can be selected in one of the modes:

  • static file attached to the script (downloaded from the Script Editor application or via the API);

  • a path formed by the arguments, which must include one of file categories;

  • a JSON list of macro paths to files can be specified as an argument.

Name: Transition

Code: transfer

Visibility condition: no

Default: — 

The component to which control is transferred if the operation completes successfully.

Name: Transition, time

Code: transferTimeout

Visibility condition: no

Default: — 

The component to which control is transferred if the HTTP response from the recognition service times out.

Name: Transition, error

Code: transferError

Visibility condition: no Default: — 

Component to which control is transferred in case of an error.